How to put AI into your small business

When AI doesn’t work out for a small business, the model is rarely the reason. More often, somebody bought a subscription before anyone decided what it was for. The team was told to “use it,” a few people poked at it for a week, and six weeks later the only visible change was one more password to remember.

This guide walks through the approach that tends to work instead. You start with work your team already does, change one piece of it, and keep an easy way to back out if it doesn’t help.

Start with a job you can describe

Before you look at any product, write down the job you want help with, in the words your team already uses. “Answer quote requests before the lead goes cold.” “Turn a phone call into a job on the board.” “Follow up on invoices more than thirty days past due.” If you can’t describe the job without naming a product, you probably haven’t found the job yet.

A well-described job has a clear start and finish, and someone on your team knows when it’s done. Skipping this step and starting from the tools is how a five-person company ends up with fifteen browser tabs open and no more hours in the week than before.

Three questions help here:

• What work is running late, gets repeated, or only lives in one person’s head?
• What does a finished version of that work look like today?
• Who gets to decide the output is good enough to go to a customer?

If you can’t answer those yet, hold off on shopping. Once you can, the first-workflow guide walks you through choosing.

Look for where the work already hurts

Spend a week paying attention before you install anything. Keep a running list on paper or in whatever notes app you already use. Whenever something is a pain, jot down the workflow, who ran into it, and whether a customer would have noticed.

You’re hoping to find repetitive work that follows a pattern. Quote requests that always need the same five questions answered. Status updates that are basically the same four sentences every time. A Monday report someone builds by copying numbers out of three different systems. That kind of work is where an AI draft can help.

Also watch for work that should stay with a person even if a draft would be easy to produce: safety calls, price exceptions, upset customers, and anything that depends on a long relationship. Mark those now so they don’t get swept into the pilot later.

Do this with the people who do the work. The owner’s picture of the week and the front desk’s picture of the week are usually quite different, and you need both.

Pick one workflow (pain × low risk)

Pick one. The best candidate is work that causes a lot of pain but wouldn’t do much damage if the AI got it wrong.

Pain means the task comes up often and someone dreads it, or it gets dropped whenever things get busy. Risk is about what happens when the draft is bad. An internal summary that a person edits before using is low risk. A customer email that goes out automatically isn’t, and neither is a number that lands in your books without anyone checking it.

It’s tempting to add a second workflow while you’re at it, because it feels efficient. Resist that. Two pilots split the limited attention you have, and you’ll get muddier answers on both. The AI Pilot Checklist is a one-page worksheet for making the choice. It covers the workflow’s name, pain, risk, baseline, time window, what success looks like, when to stop, who owns it, and notes on data.

If two workflows come out even, go with the one you can compare side by side with your current process in under two weeks. That comparison is what you’ll base the decision on, so choose the job where it’s easy to get.

Write down how it works today

Before you change anything, write down how the work goes right now, using real examples. You don’t need a dashboard. A few plain notes will do:

• How long a typical instance takes from start to finish.
• What rework looks like. Keep actual examples, like the quote that left out a line item or the intake that needed a callback.
• Where the finished work ends up.

Don’t set a target percentage you’re hoping to beat. The baseline is only a description of how things go now. Without it, every later conversation turns into competing impressions, and those are hard to make a decision from.

For a small business, five to ten examples is plenty. If the work happens so rarely that you can’t collect that many, it probably isn’t the right choice for a first pilot.

Run a shadow pilot for one to two weeks

Run the new approach alongside the old one. The old process still gets the job done, while the new one produces a draft that someone can use, edit, or ignore. Nothing goes to a customer without a person checking it first.

One to two weeks is long enough to spot patterns and short enough that people will still give you candid feedback. Pilots that drag on tend to become the new normal without anyone deciding they should. The shadow-mode guide covers how to set up the review step and what to note each day.

During the pilot, pay attention to the output more than to how excited people are about it. Enthusiasm is nice, but what you’re looking for is a draft the front desk will send after a quick edit.

Name an owner and write a short data rule

Every pilot needs one named person in charge, not a department. The owner should be able to explain any output and have the authority to switch the workflow off. If the answer is “the team,” the pilot will drift the first week someone’s out on a job site.

The data rule can be a single paragraph. It should cover what customer information is allowed into the tool, what isn’t, and where files are kept afterward, along with who’s allowed to paste in things like contracts, medical notes, card numbers, or anything about a child. For most small businesses the rule ends up short. Everyday job notes are fine to use. Payment data, passwords, and anything you wouldn’t email to a brand-new hire should stay out.

If nobody has the hours to own the pilot, that’s useful to know and nothing to be embarrassed about. It’s a good moment to look at Managed AI, because the pilot won’t run itself.

Decide whether to keep it, revise it, or stop

When the pilot window closes, make one of three calls:

• Keep. The review step held up, the owner can explain what the AI produced, and the people doing the work would rather have the draft than not.
• Revise. You picked the right job, but the handoff needs work. Change the prompt, the source files, or who reviews, and run a shorter second round. Don’t quietly widen the scope along the way.
• Stop. The risk you were worried about showed up, the drafts caused more cleanup than they saved, or nobody’s going to own this next month. Stopping still gives you something useful, because now you know where not to spend the next quarter.

Write the decision down in a few sentences somewhere the team can see it. Decisions that only live in the owner’s head tend to get reopened every time a new tool comes out.

Expand once the first one is a keep

Expanding means taking on a second workflow, or giving the same workflow to more people, after the first one is a keep. It doesn’t mean buying seats for people who weren’t in the pilot. Buying licenses first and hoping a habit forms later is one of the most common ways this goes wrong, and the measurement guide on what to count without fake ROI goes into it further.

At some point a workflow you’ve kept may need to remember things between sessions, involve more than one person, or show up on a screen your team already uses every day. That’s when the question becomes what your overall setup should look like. LaunchpadOS and the plans that include it are on arcasya.ai. We don’t list prices here. If you’d like someone to go through your week with you before choosing a plan, book a Free AI evaluation and bring your completed checklist, so the call can start from the workflow you’ve already picked.